www.c3dc.fr Pier erre Al Alliez ez, Inria Sophia Antipolis Antho - - PowerPoint PPT Presentation

A cloud computing platform for 3D scanning, documentation, preservation and dissemination of cultural heritage.

www.c3dc.fr

Pier erre Al Alliez ez, Inria Sophia Antipolis Antho hony ny Pa Pamart rt, MAP/CNRS

Partners

Culture 3D Cloud

An i ima mage ge-base sed m model deling c g cloud ud co comput puting w web s eb ser ervice : :

• Dedicated to CH community
• Versatile (scale, typologies, density)
• Provide high-density and accurate output
• Open-source based (MicMac, CGAL,…)
• User-friendly

Context

Limits o s of cur urren ent pr practices

• Requires highly specialized skills (still true for data acquisition)
• Heterogeneous results depending on software solutions

and practices (trial and error)

• Challeng

enge: facilitate adoption of 3D digitization for routine practice

Remondino, Fabio, et al. "State of the art in high density image matching." The Photogrammetric Record 29.146 (2014): 144-166.

Objectives

Digiti tizati tion : : toward d democrati tizati ation

• Large use of digital camera
• Widespread expert knowledge in image-based modeling
• Enabling non-expert end-users to perform 3D digitization
• 1. Acquisition

settings and protocols

• 2. Automatic

remote computing

• 3. Storage and

sharing

• 4. Online

visualization

Culture 3D Cloud

Cloud c ud computing ng : :

• Digitization: extraordinary computing power

(thousands of CPUs)

• Storage: continuously growing containers
• Diffusion : multi-support
• Host in TGIR HumaNum

4 simultaneous users (8cores2.7Ghz/64GbRAM/80Gb)

Scope

In progress Implemented

Platform

3D Digitization [WP l

P lead ader: r: Liv ivio io de L Luca]

3D Digitization

Settings :

• EXIF
• Exposure
• White-balance

Acquisition :

• Protocol
• Overlap
• Minimum

requirements

Processing (MicMac) :

• Best and robust

strategy Dataset:

• Simple
• Complex

Type :

• Linear
• Circular
• Random

A Modular Pipeline

• Modular

(updated and evolutive)

• Adaptive

(specificities of dataset)

• Robust

(optimization and auto-correction)

• Detection of file extension
• Adaptation to image size
• Adaptation to dataset size
• Presets
• Demanding calibration model
• Automatic initial calibration
• Robust alignment

Dense matching mode:

• Epipolar (finest)
• Multistereo (faster)
• Density:
• High (1pt/ 4px)
• Medium (1pt/ 16px)
• Low (1pt/ 64px)

Automatic data processing

6

• Strategy:
• Simple
• Random
• Stereo
• Medium
• 2 ,5 M points
• PLY : 6 7 MB

Examples

6

• Strategy:
• Complex
• Circular
• Stereo
• Medium
• 1 4 ,2 M points
• PLY : 3 6 8 MB

26

Examples

6

• Strategy:
• Complex
• Random
• Stereo
• High
• 5 M points
• PLY : 1 4 2 MB

25

Examples

6

• Strategy:
• Complex
• Circular
• Multiview
• Medium
• 5 ,3 M points
• PLY : 1 4 5 MB

56

Examples

Surface Reconstruction

16

Input Output

Dense 3D point set Colors Surface mesh

Pre -pro c e ssing Re c o nstruc tio n Po st-pro c e ssing

Design Choices

Generic & modular -> Open source libraries

17

Design Choices

Interoperability

18

Example Use Case

6

Protocol:

• Complex
• Circular
• Stereo
• Medium
• 14,2M de

de points

• PLY :

: 368m 368mo

26

Raw point set 7M points

After denoising & smoothing

Reconstructed surface 29M triangles

Hole filling

Simplified mesh 908k triangles

Simplified mesh 226 000 triangles

Simplified mesh 56 000 triangles

Simplified mesh 7 300 triangles

Conclusion

• Affordable service for non-

expert users (but data acquisition…)

• High computation ressources

accessible online

• High density and accuracy

guaranteed (if correct input data…)

• Coming soon : multifocal,

fisheye, UAV

c3dc-support@map.cnrs.fr